Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933592Ab3D3TBH (ORCPT ); Tue, 30 Apr 2013 15:01:07 -0400 Received: from mga09.intel.com ([134.134.136.24]:35039 "EHLO mga09.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933573Ab3D3TA6 (ORCPT ); Tue, 30 Apr 2013 15:00:58 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="4.87,582,1363158000"; d="scan'208";a="329098267" Subject: Re: [PATCH 1/2] Make the batch size of the percpu_counter configurable From: Tim Chen To: Christoph Lameter Cc: Andrew Morton , Tejun Heo , Al Viro , Dave Hansen , Andi Kleen , linux-kernel , linux-mm In-Reply-To: <0000013e5c32f7fd-b4bf1b22-7924-42b5-b835-eb2b5926bbf6-000000@email.amazonses.com> References: <0000013e5b24d2c5-9b899862-e2fd-4413-8094-4f1e5a0c0f62-000000@email.amazonses.com> <1367339009.27102.174.camel@schen9-DESK> <0000013e5bfd1548-a6ef7962-7b00-495b-8e83-d7a08413e165-000000@email.amazonses.com> <1367344094.27102.182.camel@schen9-DESK> <0000013e5c1377c5-49a8fca5-eb04-4e3a-a507-ce3a47fea685-000000@email.amazonses.com> <1367344522.27102.184.camel@schen9-DESK> <0000013e5c32f7fd-b4bf1b22-7924-42b5-b835-eb2b5926bbf6-000000@email.amazonses.com> Content-Type: text/plain; charset="UTF-8" Date: Tue, 30 Apr 2013 12:00:57 -0700 Message-ID: <1367348457.27102.197.camel@schen9-DESK> Mime-Version: 1.0 X-Mailer: Evolution 2.32.3 (2.32.3-1.fc14) Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 1427 Lines: 35 On Tue, 2013-04-30 at 18:27 +0000, Christoph Lameter wrote: > On Tue, 30 Apr 2013, Tim Chen wrote: > > > > But you will have to dereference the pointer whenever you want the batch > > > size from the hot path. Looks like it would be better to put the value > > > there directly. You have a list of percpu counters that can be traversed > > > to change the batch size. > > > > > > > I have considered that. But the list is not available unless we have > > CONFIG_HOTPLUG_CPU compiled in. > > percpu counters are performance sensitive and with the pointer you > will need to reference another one increasing the cache footprint. You are > touching an additional cacheline somewhere in memory frequently. Not good. > Seems like there's objection to make batch a pointer and I'll have to store its value in per cpu pointer. Will it be acceptable if I make the per cpu counter list under CONFIG_CPU_HOTPLUG default? I will need the list to go through all counters to update the batch value sizes. The alternative will be to make the configurable batch option only available under CONFIG_HOTPLUG_CPU. Which is the preferable approach or there are other clever alternatives? Tim -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/